Posted by
matlabbe on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Info-on-some-parameters-tp4568p4593.html
Hi Manuel,
For Proximity Detection theory, see section 3.3 of the paper "
Long-term online multi-session graph-based SPLAM with memory management".
1) When assembling the laser scans of a path, we have the choice of using the already optimized poses from the current graph of the map (so including loop closure detections), or using raw/optimized odometry poses only (default).
2) For debugging purpose, scan ids used for proximity detections are saved in the link computed. In database viewer, we can then re-assemble the merged scans of the proximity links for visualization (in Constraints view).
3,4,5) We can define a footprint box around the robot to filter all points in it. Sometimes the camera can see the robot itself, so this can be used to filter the obstacles on the robot.
6) When set to false, all flat obstacles are considered as ground. If true, only the biggest plane ans those at the same height are segmented as ground. For example, the top of a table would be segmented as obstacle if the camera can also see the ground. When Grid/FromDepth=true or if the laser scan is 3D, yes the obstacles are detected in 3D.
7) Experimental if false, we can compute motion transformation matching features from frame A to frame B, and then compute also motion from frame B to frame A and merge the two transforms. Not used.
8) This is what it is said, motion estimation is refined using bundle adjustment (currently only g2o is supported).
9) This will change slightly how features are matched. For Frame-To-Frame estimation, this parameter will not have a big effect. For Frame-to-Map estimation (like used by OdometryF2M), we will have better correspondences when this parameter is false. When many points in the map are projected around the same feature, we will only have one match for that feature. When parameter is true and if feature is alone in the matching radius where multiple map points are projected in it, that feature would be matched to many map points.
cheers,
Mathieu